簡介：碩士學位論文論文作者學科專業(yè)指導教師培養(yǎng)單位培養(yǎng)類別完成時間沈陽師范大學學位評定委員會單位代碼單位代碼10166單位代碼單位代碼10166大學生的學習動機與英語口語熟練水平的相關大學生的學習動機與英語口語熟練水平的相關性研究性研究THECORRELATIONBETWEENCHINESECOLLEGETHECORRELATIONBETWEENCHINESECOLLEGESTUDENTSSTUDENTS’LEARNINGMOTIVATIONANDSPOKENLEARNINGMOTIVATIONANDSPOKENENENGLISHPROFICIENCYGLISHPROFICIENCY李睿琛課程與教學論（英語）劉淑華大學外語教學部2015年05月27日全日制ITHISTHESISWASACCEPTEDASMEETINGTHERESEARCHREQUIREMENTFORTHEMASTER’SDEGREEAPPROVEDDATENAMEOFADVISORDATENAMEOFFIRSTREADERDATENAMEOFSECONDREADERAPPROVEDDATENAMEOFDEANOFDEPARTMENTOFFOREIGNLANGUAGESDATENAMEOFCHAIROFORALDEFENSECOMMITTEE

簡介：論文題目大學生心理資本問卷編制及其相關因素研究學科專業(yè)應用心理學申請人周利霞導師樊琪教授摘要積極心理學和積極組織行為學催化了心理資本及其理論的誕生，心理資本關注人的積極心理素質和潛能，越來越引起研究者的重視。以往對心理資本的研究主要集中在組織層面，而大學生作為企業(yè)人力資源新生力量的主要來源，更應該受到高度關注。本研究以大學生包括本科生和研究生為研究對象，根據(jù)國內外心理資本的相關理論，對大學生心理資本進行了研究，編制了大學生心理資本問卷，初步分析了大學生心理資本的現(xiàn)狀和差異狀況，同時探討了大學生心理資本與應對方式、職業(yè)決策自我效能感的關系，取得了初步的研究結果。本研究共調查了1175名在校本科生和研究生，使用SPSSLL．5和LIEREL8．5統(tǒng)計軟件進行數(shù)據(jù)分析，研究發(fā)現(xiàn)1、大學生心理資本包括六個維度，分別為愿景、自信、合作、樂觀、韌性和感恩，所編制的大學生心理資本問卷具有良好的信度和效度，符合測量學的標準。2、所調查大學生的心理資本及其各維度均分都高于一般水平，其中感恩得分最高，樂觀得分最低，大學生心理資本的狀況良好。3、大學生心理資本及其各維度在人口學變量上存在差異，尤其是在是否擔任學生干部、家庭成員關系、人際關系、學習成績、學校類型變量上差異非常顯著。4、大學生心理資本及其各維度與積極應對方式呈正相關，與消極應對方式呈負相關。5、大學生心理資本及其各維度與職業(yè)決策自我效能感及其各維度呈顯著正相關，大學生心理資本對職業(yè)決策自我效能感存在顯著的預測作用。本研究將心理資本理論的研究擴展到大學生群體，在全國范圍內取樣有助于了解我國大學生的心理資本的普遍現(xiàn)狀，并根據(jù)研究結果進行了一系列的討論，有助于提高高校和教育者對大學生心理資本的重視，推動學校心理教育的發(fā)展，同時，本文也豐富了心理資本理論和研究，具有理論和現(xiàn)實意義。關鍵詞心理資本；大學生；應對方式；職業(yè)決策自我效能感IIIIIIILFLLLLLRLLRIIIIIY2096030PSYCHOLOGICALCAPITALISPOSITIVE．3THECOLLEGESTUDENTS’PSYCHOLOGICALCAPITALANDITSDIMENSIONDIFFERINSOMEDEMOGRAPHICVARIABLES；ESPECIALLYTHEDIFFERENCESONWHETHERJOINEDSTUDENTS’UNION,FAMILYMEMBERS’RELATIONSHIPS，INTERPERSONALRELATIONSHIPS，ACADEMICPERFORMANCE，SCHOOLTYPESAREVERYSIGNIFICANT．4PSYCHOLOGICALCAPITALINCLUDINGITSDIMENSIONSALEPOSITIVELYCORRELATEDWITHPOSITIVECOPINGSTYLE，ANDNEGATIVELYCORRELATEDWITHTHENEGATIVECOPINGSTYLE．5THECOLLEGESTUDENTS’PSYCHOLOGICALCAPITALINCLUDINGITSDIMENSIONSANDCAREERDECISIONMAKINGSELFEFFICACYINCLUDINGITSDIMENSIONSARESIGNIFICANTLYCORRELATED；PSYCHOLOGICALCAPITALHASASIGNIFICANTPREDICTIVEEFFECTONCAREERDECISIONMAKINGSELFEFFICACY．THISSTUDYEXTENDSPSYCHOLOGICALCAPITALRESEARCHTOCOLLEGESTUDENTS，NATIONWIDESAMPLINGHELPSTOKNOWTHEGENERALLYSTATUSOFPSYCHOLOGICALCAPITALOFCHINESECOLLEGESTUDENTS，ANDASERIESOFDISCUSSIONSBASEDONTHERESULTSWOULDHELPTOENHANCETHEATTENTIONOFUNIVERSITIESANDEDUCATORSONCOLLEGESTUDENTS’PSYCHOLOGICALCAPITAL，ANDALSOIT’SHELPFULTOPROMOTETHEDEVELOPMENTOFPSYCHOLOGICALEDUCATIONINUNIVERSITIES．WHAT’SMORE，THESTUDYENRICHEDPSYCHOLOGICALCAPITAL’STHEORYANDRESEARCH，ITHASTHEORETICALANDPRACTICALSIGNIFICANCE．KEYWORDSPSYCHOLOGICALCAPITAL；COLLEGESTUDENTS；COPINGSTYLE；CAREERDECISIONMAKINGSELFEFFICACYIII

簡介：學號2007020164研究生姓名李莉聯(lián)系電話15866760373EMAIL598797173QQCOM所在院系心理學院

簡介：大學生呼吸肌耐力與肺活量、最大攝氧量的相關研究論文作者簽名曼紐莖曼指導教師簽名論文評閱人1評閱人2評閱人3評閱人4評閱人5答辯委員會主席委員1委員2委員3委員4委員5答辯日期瑚厶∈邁致謝時光如梭，轉眼間兩年的碩士學習生活即將過去，在這段難忘的日子里，曾得到許多老師、同學、朋友及家人的關心幫助，作者謹以本文獻給他們，以此表達深深地祝福和誠摯的謝意。首先感謝我的導師徐玉明教授。徐老師知識淵博、學術嚴謹。他忘我的工作精神、誠摯的處事態(tài)度、風趣幽默的語言、細致入微的關懷給我留下了深刻的印象，使我受益匪淺。從論文開題、指標選擇、測試設計到最后的撰寫論文，徐老師都給與了極大的幫助和關懷，在此致以衷心的敬意和深深的謝意感謝我的同F(xiàn)J師兄弟陳玉平、羅宇婷、鄭冬霞、張亞新、喬暢等，感謝你們對我學7工作上的幫助，正是有了你們的幫助和鼓勵，我才能順利完成實驗，在此對你們表示深深的謝意。感謝同班好友王亞圣、劉晴、李征等，在我碩士期間陪我度過的許多美好時光，每當想起這些，我的心里充滿溫暖和熱情，在這里祝你們前程似錦，幸福美好。感謝我的家人，尤其是我的父親和母親二十多年來給與我的大力支持和無私的愛最后，向在百忙中抽出時間參與我論文答辯和評審的各位專家學者表示由衷的感謝

簡介：分類號密級無UDC單位代碼10118山西師范大學研究生碩士學位論文大學生體育鍛煉與情緒管理能力的相關性研究屈振榮屈振榮指導教師張海潮教授山西師范大學申請學位級別教育學碩士專業(yè)名稱體育教育訓練學論文提交日期2015年4月7日論文答辯日期2015年5月24日學位授予單位山西師范大學學位授予日期2015年6月4日答辯委員會主席邱招義教授評閱人張廣林教授崔浩瀾教授2015年5月20日中文摘要I論文題目大學生體育鍛煉與情緒管理能力的相關性研究專業(yè)體育教育訓練學碩士生屈振榮簽名指導教師張海潮簽名摘要生活滿意感和生命意義感是衡量大千世界中個體生活質量的綜合性心理指標，其中對生活滿意感和生命意義感評定的重要因素就是對情緒管理能力的控制和調節(jié)。因而，提高生活滿意感和生命意義感的總體水平，加強情緒管理的控制，對于提高人們的生活質量就顯得尤為重要。當代大學生具有心理變化大，心理素質不穩(wěn)定，可塑性較強的心理特征，受社會、周圍環(huán)境等各方面影響和自身心理特點的制約，易產生多重負面心理體驗，如空虛、寂寞、迷茫等，處于心理亞健康狀態(tài)，為各種心理問題埋下隱患。因此，重視當代大學生的生活滿意感、生命意義感，提高大學生的情緒管理能力，通過體育鍛煉的方式達到消除各種消極的心理影響的效果，是預防大學生心理疾病，優(yōu)化大學生心理健康的需要?？茖W的體育鍛煉有利于調節(jié)不良情緒，增進積極情感，促進人的身心健康。但是當代大學生對體育鍛煉強身健心功效的認識程度并不深刻，其治療心理亞健康，預防心理疾病的心理學功效并未得到廣泛認識和重視，相關研究論證也十分欠缺。因此，本論文所要研究的核心內容是通過探討大學生參與體育鍛煉與情緒管理能力之間的相互關系，為改善大學生的生活質量，提高大學生的情緒管理能力，優(yōu)化大學生的心理健康提供理論參考，為普通高校制定大學生情緒管理能力和心理健康教育方案提供有益的借鑒，為體育鍛煉的強身健心功效提供實證研究資料。采用大學生情緒管理能力問卷、體育活動等級量表和體育鍛煉情況調查表。問卷信、效度較高，對山西省五所普通高等學校的1000名在校大學生進行調查，運用SPSS170數(shù)據(jù)統(tǒng)計軟件進行數(shù)據(jù)分析。結論如下（1）大學生參加體育鍛煉的意識較好，但達到標準體育鍛煉人口的人數(shù)較少，即參加體育鍛煉的運動量不太理想。（2）體育鍛煉的時間、強度、頻率與情緒管理能力均存在顯著相關性，且體育鍛煉強度和頻率與情緒管理能力的相關系數(shù)較高?！娟P鍵詞】大學生體育鍛煉情緒管理能力相關性研究【論文類型】關系型研究

簡介：AI遼SEARCHONTHEI迮LATIONBETWEENBODYIMAGEANDWEIGHTLOSSBEHAVIOROFCOLLEGESTUDENTSADISSERTATIONSUBMITTEDTOSOUTHEASTUNIVERSITYFORTHEACADEMICDEGREEOFMASTEROFEDUCATIONBYZHANGYANSUPERVISEDBYSUPERVISEDBYPROF．HELUNSCHOOLOFHUMANITIESSOUTHEASTUNIVERSITYDECEMBER2009摘要摘要所謂身體意象，是指對自己身體的意象，是關于自己的外貌、身體型象在他人面前是怎樣的一種想象，也包括別人是如何看待自己的身體功能方面的意象。也就是說，意象一方面是自己對自己身體的想象，另一方面，是別人對自己身體的想象。大學生正處于青春期，是自我概念發(fā)展的關鍵期，正面臨著自我同一性與發(fā)展危機，在這一階段，個體開始關注自我，其自我意識迅速發(fā)展并逐步形成。這一時期急劇的身體成熟，使得他們更加注意關注自己的身體，身體滿意度逐漸下降，減肥行為也隨之增加，他們成為追求苗條體型、追求時尚的主要人群之一。傳統(tǒng)觀念的人們，大多數(shù)會認為肥胖者們才會去減肥，殊不知在如今大學生中存在很大～部分群體，盡管在別人的眼里他們的體重正?；蚱?，但他們仍然還存在減肥行為，而這些與自身體重、體型根本無關。因此，本文從大學生著手，探討他們的減肥行為，力圖探索這一群體的身體意象與減肥行為的關系，分析其減肥行為背后的心理動因。本研究采用采取整群抽樣的方法，各年級抽取兩個班的學生進行檢查。以STORZ．GREENE體型圖、多維自我體象關系問卷MBSRQ和自編減肥行為問卷為工具對766名大學生進行了調查，得出結論L、大多數(shù)學生對于他們的身體意象是滿意，只有少數(shù)學生不滿意，甚至存在負性扭曲現(xiàn)象。大學生身體意象在性別、年級、專業(yè)、生源及BML分組方面存在差異。2、總體看來，大部分學生傾向于選擇中間范圍的體重作為自己實際體重和理想體重。這些認知在性別、年級、專業(yè)、生源及BML分組方面存在差異。3、大學生存在減肥行為，女生表現(xiàn)出更高的減肥意向與關注；在減肥群體中，大多是正常體重的群體，而且很多都是心理認同的肥胖，和生理指數(shù)關系不大。4、大學生身體意象與減肥行為存在顯著負相關，越是低程度的身體意象，減肥意向和減肥關注程度越高。5、大學生對自己身體的不滿意和對體重的認知錯覺是導致大學生減肥行為的重要原因之關鍵詞；大學生身體意象減肥行為

簡介：碩士學位論文大一學生網絡成癮與人格特質、孤獨、自尊及學校適應性的相關性研究ONRELEVANCEOFFIRSTYEARCOLLEGESTUDENTS’INTERNETADDICTIONTOPERSONALTRAITS，LONELINESS，SELFESTEEMANDCOLLEGEADJUSTMENT作者齊曉穎指導教師學科專業(yè)研究方向學位類型姜英杰副教授’發(fā)展與教育心理學網絡成癮‘高校教師東北師范大學學位評定委員會2010年5月獨創(chuàng)性聲明本人聲明所呈交的學位論文是本人在導師指導下進行的研究工作及取得的研究成果。據(jù)我所知，除了文中特LLIJL以標注和致謝的地方外，論文中不包含其他人已經發(fā)表或撰寫過的研究成果，也不包含為獲得東北師范大學或其他教育機構的學位或證書而使用過的材料。與我一同工作的同志對本研究所做的任何貢獻均已在論文中作了明確的說明并表示謝意。‘學位論文作者簽名學位論文版權使用授權書本學位論文作者完全了解東北師范大學有關保留、使用學位論文的規(guī)定，即東北師范大學有權保留并向國家有關部門或機構送交學位論文的復印件和磁盤，允許論文被查閱和借閱。本人授權東北師范大學可以將學位論文的全部或部分內容編入有關數(shù)據(jù)庫進行檢索，可以采用影印、縮印或其它復制手段保存、匯編學位論文。保密的學位論文在解密后適用本授權書學位論文作者簽名壺隆題指導教師簽名日期日期學位論文作者畢業(yè)后去向電話郵編13944691028132101

簡介：四川師范大學學位論文獨創(chuàng)性聲明本人聲明所呈交學位論文太眾佳媒佳直信星懟丕回眭別太堂生佳直叢翅髭響的扭差比較巫究二叢四叫垣范太堂趁倒，是本人在導師龐面主教援指導下，獨立進行研究工作所取得的成果。除文中已經注明引用的內容外，本論文不含任何其他個人或集體己經發(fā)表或撰寫過的作品或成果。對本文的研究做出重要貢獻的個人和集體，均已在文中以明確方式標明。本聲明的法律結果由本人承擔。本人承諾己提交的學位論文電子版與論文紙本的內容一致。如因不符而引起的學術聲譽上的損失由本人自負。學位論文作者稀，『13芬及簽字日期2．O／B年6月多’日四川師范大學學位論文版權使用授權書本人同意所撰寫學位論文的使用授權遵照學校的管理規(guī)定學校作為申請學位的條件之一，學位論文著作權擁有者須授權所在大學擁有學位論文的部分使用權，即1己獲學位的研究生必須按學校規(guī)定提交印刷版和電子版學位論文，可以將學位論文的全部或部分內容編入有關數(shù)據(jù)庫供檢索；2為教學、科研和學術交流目的，學?？梢詫⒐_的學位論文或解密后的學位論文作為資料在圖書館、資料室等場所或在有關網絡上供閱讀、瀏覽。本人授權萬方數(shù)據(jù)電子出版社將本學位論文收錄到中國學位論文全文數(shù)據(jù)庫，并通過網絡向社會公眾提供信息服務。同意按相關規(guī)定享受相關權益。保密的學位論文在解密后適用本授權書學位論文作者簽名私LF私簽字日期加佑年多月≯≥日翩魏寵簽字日期勃鰣年莎月≥JJL日萬方數(shù)據(jù)四川師范大學碩士學位論文II萬方數(shù)據(jù)

簡介：??????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????“?ABC1DEF??BG???HIJKLMN?

簡介：獨創(chuàng)性聲明本人聲明所呈交的學位論文是本人在導師指導下進行的研究工作及取得的研究成果。論文中除了特別加以標注和致謝的地方外，不包含其他人已經發(fā)表或撰寫過的研究成果，也不包含為獲得其他教育機構的學位或證書而使用過的材料。與我一同工作的同志對本研究所做的任何貢獻均己在論文中作了明確的說明并表示謝意。學位論文作者簽名．I琵釋簽字日期護易年鄉(xiāng)月弓P日學位論文版權使用授權書本學位論文作者完全了解J匕塞跡獨醫(yī)堂隨有關保存、使用學位論文的管理辦法。有權保留并向國家有關部門或機構送交論文的復印件和磁盤，允許論文被查閱和借閱。本人授權JE塞垃獨醫(yī)堂醫(yī)可以將學位論文的全部或部分內容編入有關數(shù)據(jù)庫進行檢索，可以采用影印、縮印或掃描等復制手段保存、匯編學位論文。保密的學位論文在解密后適用本授權書學位論文作者虢琵智導師虢簽字日期2．O16年鄉(xiāng)月；P日簽字日期砂F6年學位論文作者畢業(yè)后去向工作單位通訊地址電話郵編．B1寧欠≥月字口日平會對論文質量產生顯著影響，因此認為增加與發(fā)達國家的合著是提升合著論文質量的途徑之一；5基金資助情況會對論文質量產生顯著正向影響，提示可以通過科研項目之間的學術交流與合作，提高科研產出的質量；6相對于傳統(tǒng)統(tǒng)計方法而言，結構方程模型考慮了多個因素的協(xié)同作用，同時考慮了因變量的協(xié)同效應，避免了犯二類錯誤，評價結果更為客觀。同時本研究也存在一定的不足，研究將在多個醫(yī)學研究領域進行實證分析，確保研究結果在醫(yī)學領域的適用性，同時將進一步通過專家咨詢，確定各指標對論文質量的影響系數(shù)。關鍵詞國際合著；論文質量；相關分析；結構方程

簡介：STABLEDETERMINISTICMULTITHREADINGTHROUGHSCHEDULEMEMOIZATIONHEMINGCUI,JINGYUEWU,CHIACHETSAI,JUNFENGYANG{HEMING,JINGYUE,CT2459,JUNFENG}CSCOLUMBIAEDUCOMPUTERSCIENCEDEPARTMENTCOLUMBIAUNIVERSITYNEWYORK,NY10027ABSTRACTADETERMINISTICMULTITHREADINGDMTSYSTEMELIMINATESNONDETERMINISMINTHREADSCHEDULING,SIMPLIFYINGTHEDEVELOPMENTOFMULTITHREADEDPROGRAMSHOWEVER,EXISTINGDMTSYSTEMSAREUNSTABLETHEYMAYFORCEAPROGRAMTOADVENTUREINTOVASTLYDIFFERENTSCHEDULESEVENFORSLIGHTLYDIFFERENTINPUTSOREXECUTIONENVIRONMENTS,DEFEATINGMANYBENEFITSOFDETERMINISMMOREOVER,FEWEXISTINGDMTSYSTEMSWORKWITHSERVERPROGRAMSWHOSEINPUTSARRIVECONTINUOUSLYANDNONDETERMINISTICALLYTERNISASTABLEDMTSYSTEMTHEKEYNOVELTYINTERNISTHEIDEAOFSCHEDULEMEMOIZATIONTHATMEMOIZESPASTWORKINGSCHEDULESANDREUSESTHEMONFUTUREINPUTS,MAKINGPROGRAMBEHAVIORSSTABLEACROSSDIFFERENTINPUTSASECONDNOVELTYINTERNISTHEIDEAOFWINDOWINGTHATEXTENDSSCHEDULEMEMOIZATIONTOSERVERPROGRAMSBYSPLITTINGCONTINUOUSREQUESTSTREAMSINTOWINDOWSOFREQUESTSOURTERNIMPLEMENTATIONRUNSONLINUXITOPERATESASUSERSPACESCHEDULERS,REQUIRINGNOCHANGESTOTHEOSANDONLYAFEWLINESOFCHANGESTOTHEAPPLICATIONPROGRAMSWEEVALUATEDTERNONADIVERSESETOF14PROGRAMSEG,APACHEANDMYSQLWITHREALANDSYNTHETICWORKLOADSOURRESULTSSHOWTHATTERNISEASYTOUSE,MAKESPROGRAMSMOREDETERMINISTICANDSTABLE,ANDHASREASONABLEOVERHEAD1INTRODUCTIONMULTITHREADEDPROGRAMSAREDIFFICULTTOWRITE,TEST,ANDDEBUGAKEYREASONISNONDETERMINISMDIFFERENTRUNSOFAMULTITHREADEDPROGRAMMAYSHOWDIFFERENTBEHAVIORS,DEPENDINGONHOWTHETHREADSINTERLEAVE35TWOMAINFACTORSMAKETHREADSINTERLEAVENONDETERMINISTICALLYTHEFIRSTISSCHEDULING,HOWTHEOSANDHARDWARESCHEDULETHREADSSCHEDULINGNONDETERMINISMISNOTESSENTIALANDCANBEELIMINATEDWITHOUTAFFECTINGCORRECTNESSFORMOSTPROGRAMSTHESECONDISINPUT,WHATDATAINPUTDATAARRIVESATWHATTIMEINPUTTIMINGINPUTNONDETERMINISMSOMETIMESISESSENTIALBECAUSEMAJORCHANGESININPUTSREQUIREDIFFERENTSCHEDULESHOWEVER,FREQUENTLYINPUTNONDETERMINISMISNOTESSENTIALANDTHESAMESCHEDULECANBEUSEDTOPROCESSMANYDIFFERENTINPUTS§22WEBELIEVENONESSENTIALNONDETERMINISMSHOULDBEELIMINATEDINFAVOROFDETERMINISMDETERMINISTICMULTITHREADINGDMTSYSTEMS13,22,41MAKETHREADSMOREDETERMINISTICBYELIMINATINGSCHEDULINGNONDETERMINISMSPECIFICALLY,THEYCONSTRAINAMULTITHREADEDPROGRAMSUCHTHATITALWAYSUSESTHESAMETHREADSCHEDULEFORTHESAMEINPUTBYDOINGSO,THESESYSTEMSMAKEPROGRAMBEHAVIORSREPEATABLE,INCREASETESTINGCONFIDENCE,ANDEASEBUGREPRODUCTIONUNFORTUNATELY,THOUGHEXISTINGDMTSYSTEMSELIMINATESCHEDULINGNONDETERMINISM,THEYDONOTREDUCEINPUTNONDETERMINISMINFACT,THEYMAYAGGRAVATETHEEFFECTSOFINPUTNONDETERMINISMBECAUSEOFTHEIRDESIGNLIMITATIONWHENSCHEDULINGTHETHREADSTOPROCESSANINPUT,THEYCONSIDERONLYTHISINPUTANDIGNOREPREVIOUSSIMILARINPUTSTHISSTATELESSDESIGNMAKESSCHEDULESOVERDEPENDENTONINPUTS,SOTHATASLIGHTCHANGETOINPUTSMAYFORCEAPROGRAMTOADVENTUREINTOAVASTLYDIFFERENT,POTENTIALLYBUGGYSCHEDULE,DEFEATINGMANYBENEFITSOFDETERMINISMWECALLTHISTHEINSTABILITYPROBLEMTHISPROBLEMISCONFIRMEDBYOURRESULTS§821FROMANEXISTINGDMTSYSTEM13INFACT,EVENWITHTHESAMEINPUT,EXISTINGDMTSYSTEMSMAYSTILLFORCEAPROGRAMINTODIFFERENTSCHEDULESFORMINORCHANGESINTHEEXECUTIONENVIRONMENTSUCHASPROCESSORTYPEANDSHAREDLIBRARYTHUS,DEVELOPERSMAYNOLONGERBEABLETOREPRODUCEBUGSBYRUNNINGTHEIRPROGRAMONTHEBUGINDUCINGINPUT,BECAUSETHEIRMACHINEMAYDIFFERFROMTHEMACHINEWHERETHEBUGOCCURREDTHISPAPERPRESENTSTERN,ASCHEDULECENTRIC,STATEFULDMTSYSTEMITADDRESSESTHEINSTABILITYPROBLEMUSINGANIDEACALLEDSCHEDULEMEMOIZATIONTHATMEMOIZESPASTWORKINGSCHEDULESANDREUSESTHEMFORFUTUREINPUTSSPECIFICALLY,TERNMAINTAINSACACHEOFPASTSCHEDULESANDTHEINPUTCONSTRAINTSREQUIREDTOREUSETHESESCHEDULESWHENANINPUTARRIVES,TERNCHECKSTHEINPUTAGAINSTTHEMEMOIZEDCONSTRAINTSFORACOMPATIBLESCHEDOURCHOICEOFSCHEDULEREPRESENTATIONINTERN§22,ANDWHYWECANREUSESCHEDULESACROSSINPUTS§2321THEINSTABILITYPROBLEMADMTSYSTEMIS,CONCEPTUALLY,AFUNCTIONTHATMAPSANINPUTITOASCHEDULESTHEPROPERTIESOFTHISFUNCTIONARETHATTHESAMEISHOULDMAPTOTHESAMESANDTHATSISAFEASIBLESCHEDULEFORPROCESSINGIASTABLEDMTSYSTEMSUCHASTERNHASANADDITIONALPROPERTYITMAPSSIMILARINPUTSTOTHESAMESCHEDULEEXISTINGDMTSYSTEMS,HOWEVER,TENDTOMAPSIMILARINPUTSTODIFFERENTSCHEDULES,THUSSUFFERINGFROMTHEINSTABILITYPROBLEMWEARGUETHATTHISPROBLEMISINHERENTINEXISTINGDMTSYSTEMSBECAUSETHEYARESTATELESSTHEYMUSTPROVIDETHESAMESCHEDULEFORANINPUTACROSSDIFFERENTRUNS,USINGINFORMATIONONLYFROMTHECURRENTRUNTOFORCETHREADSTOCOMMUNICATEEG,ACQUIRELOCKSORACCESSSHAREDMEMORYDETERMINISTICALLY,EXISTINGDMTSYSTEMSCANNOTRELYONPHYSICALCLOCKSINSTEAD,THEYMAINTAINALOGICALCLOCKPERTHREADTHATTICKSDETERMINISTICALLYBASEDONTHECODETHISTHREADHASRUNMOREOVER,THREADSMAYCOMMUNICATEONLYWHENTHEIRLOGICALCLOCKSHAVEDETERMINISTICVALUESEG,SMALLESTACROSSTHELOGICALCLOCKSOFALLTHREADS41BYINDUCTION,LOGICALCLOCKSMAKETHREADSDETERMINISTICHOWEVER,THEPROBLEMWITHLOGICALCLOCKSISTHATFOREFFICIENCY,THEYMUSTTICKATROUGHLYTHESAMERATETOPREVENTATHREADWITHASLOWERCLOCKFROMSTARVINGOTHERSTHUS,EXISTINGDMTSYSTEMSHAVETOTIETHEIRLOGICALCLOCKSTOLOWLEVELINSTRUCTIONSEXECUTEDEG,COMPLETEDLOADS41CONSEQUENTLY,ASMALLCHANGETOTHEINPUTOREXECUTIONENVIRONMENTMAYALTERAFEWINSTRUCTIONSEXECUTED,INTURNALTERINGTHELOGICALCLOCKSANDSUBSEQUENTTHREADCOMMUNICATIONSTHATIS,ASMALLCHANGETOTHEINPUTOREXECUTIONENVIRONMENTMAYCASCADEINTOAMUCHDIFFERENTEG,CORRECTVSBUGGYSCHEDULE22SCHEDULEREPRESENTATIONANDDETERMINISMPREVIOUSDMTSYSTEMSHAVECONSIDEREDTWOTYPESOFSCHEDULES1ADETERMINISTICORDEROFSHAREDMEMORYACCESSES13,22AND2ASYNCHRONIZATIONORDERIE,ATOTALORDEROFSYNCHRONIZATIONOPERATIONS41THEFIRSTTYPEOFSCHEDULESARETRULYDETERMINISTICEVENIFTHEREARERACES,BUTTHEYARECOSTLYTOENFORCEONCOMMODITYHARDWAREEG,UPTO10TIMESOVERHEAD13THESECONDTYPECANBEEFFICIENTLYENFORCEDEG,16OVERHEAD41BECAUSEMOSTCODEISNOTSYNCHRONIZATIONCODEANDCANRUNINPARALLELHOWEVER,THEYAREDETERMINISTICONLYFORINPUTSTHATLEADTORACEFREERUNS41,46TERNREPRESENTSSCHEDULESASSYNCHRONIZATIONORDERSFOREFFICIENCYANADDITIONALBENEFITISTHATSYNCHRONIZATIONORDERSCANBEREUSEDMOREFREQUENTLYTHANMEMORYACCESSORDERSCFNEXTSUBSECTIONMOREOVER,RESEARCHERSHAVEFOUNDTHATMANYCONCURRENCYERRORSARENOTDATAPROGRAMINPUTCONSTRAINTSFORSCHEDULEREUSEPBZIP2SAMENUMBEROFFILEBLOCKSNUMBLOCKSORBANDTHREADSPAPACHEFORGROUPSOFTYPICALHTTPGETREQUESTS,SAMECACHESTATUSANDRESPONSESIZESFFTSAMENUMBEROFTHREADSPLUSAMENUMBEROFTHREADSP,SIZEOFTHEMATRIXN,ANDBLOCKSIZEBBARNESSAMENUMBEROFTHREADSNPROCANDVALUESOFVARIABLESDTIMEANDTSTOPTABLE1INPUTCONSTRAINTSOFFIVEPROGRAMSTOREUSESCHEDULESIDENTIFIERSWITHOUTADASHARECONFIGURATIONVARIABLES,ANDTHOSEWITHADASHARECOMMANDLINEOPTIONSRACES,BUTATOMICITYANDORDERVIOLATIONS39THESEERRORSCANBEDETERMINISTICALLYREPRODUCEDORAVOIDEDUSINGONLYSYNCHRONIZATIONORDERSALTHOUGHDATARACESMAYSTILLMAKERUNSWHICHREUSESCHEDULESNONDETERMINISTIC,TERNISLESSPRONETOTHISPROBLEMTHANEXISTINGDMTSYSTEMS41BECAUSEITHASTHEFLEXIBILITYTOSELECTSCHEDULESIFITDETECTSARACEINAMEMOIZEDSCHEDULE,ITCANSIMPLYDISCARDTHISSCHEDULEANDMEMOIZEANOTHERTHISSELECTIONTASKISOFTENEASYBECAUSEMOSTSCHEDULESARERACEFREEINRARECASES,TERNMAYBEUNABLETOFINDARACEFREESCHEDULE,RESULTINGINNONDETERMINISTICRUNSHOWEVER,WEARGUETHATINPUTNONDETERMINISMCANNOTBEFULLYELIMINATEDANYWAY,SOWEMAYASWELLTOLERATESOMESCHEDULINGNONDETERMINISM,FOLLOWINGTHEENDTOENDARGUMENT23WHYCANWEREUSESCHEDULESTHISSUBSECTIONPRESENTSANINTUITIVEANDANEMPIRICALARGUMENTTOSUPPORTOURINSIGHTTHATWECANFREQUENTLYREUSESCHEDULESFORMANYPROGRAMS/WORKLOADSINTUITIVELY,SYNCHRONIZATIONOPERATIONSMAPTODEVELOPERINTENTSOFINTERTHREADCONTROLFLOWBYENFORCINGTHESAMESYNCHRONIZATIONORDER,WEFIXTHESAMEINTERTHREAD“PATH,”BUTSTILLALLOWMANYDIFFERENTINPUTSTOFLOWDOWNTHISPATHTHISOBSERVATIONISSIMILARLYMADEFORSEQUENTIALPATHS11,12,26TOEMPIRICALLYVALIDATEOURINSIGHT,WESTUDIEDTHEINPUTCONSTRAINTSTOREUSESCHEDULESFORFIVEPROGRAMS,INCLUDINGAPARALLELCOMPRESSIONUTILITYPBZIP2THEAPACHEWEBSERVERANDTHREESCIENTIFICPROGRAMSFFT,LU,ANDBARNESINSPLASH2TABLE1SHOWSTHERESULTSFORALLPROGRAMSSTUDIEDWEFOUNDTHATTHEINPUTCONSTRAINTSWEREOFTENGENERAL,ALLOWINGFREQUENTREUSESOFSCHEDULESFORINSTANCE,PBZIP2CANUSETHESAMESCHEDULETOCOMPRESSMANYDIFFERENTFILES,ASLONGASTHENUMBEROFTHREADSANDTHENUMBEROFFILEBLOCKSREMAINTHESAME3OVERVIEWOURDESIGNOFTERNADHERESTOTHEFOLLOWINGGOALS

簡介：TRUSTANDPROTECTIONINTHEILLINOISBROWSEROPERATINGSYSTEMSHUOTANG,HAOHUIMAI,SAMUELTKINGUNIVERSITYOFILLINOISATURBANACHAMPAIGNABSTRACTCURRENTWEBBROWSERSARECOMPLEX,HAVEENORMOUSTRUSTEDCOMPUTINGBASES,ANDPROVIDEATTACKERSWITHEASYACCESSTOMODERNCOMPUTERSYSTEMSINTHISPAPERWEINTRODUCETHEILLINOISBROWSEROPERATINGSYSTEMIBOS,ANEWOPERATINGSYSTEMANDANEWBROWSERTHATREDUCESTHETRUSTEDCOMPUTINGBASEFORWEBBROWSERSINOURARCHITECTUREWEEXPOSEBROWSERLEVELABSTRACTIONSATTHELOWESTSOFTWARELAYER,ENABLINGUSTOREMOVEALMOSTALLTRADITIONALOSCOMPONENTSANDSERVICESFROMOURTRUSTEDCOMPUTINGBASEBYMAPPINGBROWSERABSTRACTIONSTOHARDWAREABSTRACTIONSDIRECTLYWESHOWTHATTHISARCHITECTUREISFLEXIBLEENOUGHTOENABLENEWBROWSERSECURITYPOLICIES,CANSTILLSUPPORTTRADITIONALAPPLICATIONS,ANDADDSLITTLEOVERHEADTOTHEOVERALLBROWSINGEXPERIENCE1INTRODUCTIONWEBBASEDAPPLICATIONSWEBAPPS,BROWSERS,ANDOPERATINGSYSTEMSHAVEBECOMEPOPULARTARGETSFORATTACKERSOFCOMPUTERSYSTEMSVULNERABILITIESINWEBAPPSAREWIDESPREADANDINCREASINGFOREXAMPLE,CROSSSITESCRIPTINGXSS,WHICHISEFFECTIVELYAFORMOFSCRIPTINJECTIONINTOAWEBAPP,RECENTLYOVERTOOKTHEUBIQUITOUSBUFFEROVERFLOWASTHEMOSTCOMMONSECURITYVULNERABILITY50VULNERABILITIESINWEBBROWSERSARELESSCOMMONTHANWEBAPPVULNERABILITIES,BUTSTILLOCCUROFTENFOREXAMPLE,IN2009INTERNETEXPLORER,CHROME,SAFARI,ANDFIREFOXHAD349NEWSECURITYVULNERABILITIES4,ANDATTACKERSEXPLOITBROWSERSCOMMONLY53,37,42,41,4VULNERABILITIESINLIBRARIES,SYSTEMSERVICES,ANDOPERATINGSYSTEMSARELESSCOMMONTHANVULNERABILITIESINBROWSERS,BUTARESTILLPROBLEMATICFORMODERNSYSTEMSFOREXAMPLE,GLIBC,GTK,X,ANDLINUXHAD114NEWSECURITYVULNERABILITIESIN20091,ANDIN2009THEMOSTCOMMONLYATTACKEDVULNERABILITYWASAREMOTECODEEXECUTIONBUGINTHEWINDOWSKERNEL4HOWEVER,NOTALLATTACKSONWEBAPPS,BROWSERS,ANDOPERATINGSYSTEMSAREEQUALLYVIRULENTATTHETOPOFTHECOMPUTERSTACK,ATTACKSONWEBAPPS,SUCHASXSS,OPERATEWITHINCURRENTBROWSERSECURITYPOLICIESTHATCONTAINTHEDAMAGETOTHEVULNERABLEWEBAPPMOVINGDOWNTHECOMPUTERSTACK,ATTACKSONBROWSERSCANCAUSEMOREDAMAGEBECAUSEASUCCESSFULATTACKGIVESTHEATTACKERACCESSTOBROWSERDATAFORALLWEBAPPSANDACCESSTOOTHERRESOURCESONTHESYSTEMATTHELOWESTLAYERSOFTHECOMPUTERSTACK,ATTACKSONLIBRARIES,SHAREDSYSTEMSERVICES,ANDOPERATINGSYSTEMSARETHEMOSTSERIOUSATTACKSBECAUSEATTACKERSCANACCESSARBITRARYSTATESANDEVENTS,GIVINGTHEMCOMPLETECONTROLOFTHESYSTEMOVERALL,THESETRENDSINDICATETHATVULNERABILITIESHIGHERINTHECOMPUTERSTACKAREMORECOMMON,BUTVULNERABILITIESLOWERINTHECOMPUTERSTACKPROVIDEATTACKERSWITHMORECONTROLANDAREMOREDAMAGINGINTHISPAPERWEFOCUSONPREVENTINGANDCONTAININGATTACKSONBROWSERS,LIBRARIES,SYSTEMSERVICES,ANDOPERATINGSYSTEMS–THELOWERLAYERSOFTHECOMPUTERSTACKCURRENTRESEARCHEFFORTSINTOMORESECUREWEBBROWSERSHELPIMPROVETHESECURITYOFBROWSERS,BUTREMAINSUSCEPTIBLETOATTACKSONLOWERLAYERSOFTHECOMPUTERSTACKTHEOPWEBBROWSER26,GAZELLE52,CHROME11,ANDCHROMEOS25PROPOSENEWBROWSERARCHITECTURESFORSEPARATINGTHEFUNCTIONALITYOFTHEBROWSERFROMSECURITYMECHANISMSANDPOLICIESHOWEVER,THESEMORESECUREWEBBROWSERSAREALLBUILTONTOPOFCOMMODITYOPERATINGSYSTEMSANDINCLUDECOMPLEXUSERMODELIBRARIESANDSHAREDSYSTEMSERVICESWITHINTHEIRTRUSTEDCOMPUTINGBASETCBEVENKERNELDESIGNSWITHSTRONGISOLATIONBETWEENOSCOMPONENTSEG,MICROKERNELS24,27,28ANDINFORMATIONFLOWKERNELS18,57,33STILLHAVEOSSERVICESTHATARESHAREDBYALLAPPLICATIONS,WHICHATTACKERSCANCOMPROMISEANDSTILLCAUSEDAMAGEHEREAREAFEWWAYSTHATANATTACKERCANSTILLCAUSEDAMAGETOMORESECUREWEBBROWSERSBUILTONTOPOFTRADITIONALOSES1HARDWAREBROWSERABSTRACTIONSIBOSKERNELL4REFERENCEMONITORUISTORAGEWEBPAGEINSTANCEWEBPAGEINSTANCEWEBPAGEINSTANCEMAPSCOMBINGCOMUIUCEDUUNIXLAYERTRADITIONALAPPLICATIONSLOCALHOSTMOUSE/KBDDRIVERNICDRIVERNETMANAGERNICMOUSE/KBDVIDEONETPROCESSESFIGURE1OVERALLIBOSARCHITECTUREOURSYSTEMCONTAINSUSERMODEDRIVERS,BROWSERSAPIMANAGERS,WEBPAGEINSTANCES,ANDTRADITIONALPROCESSESTOMANAGETHEINTERACTIONSBETWEENTHESECOMPONENTS,WEUSEAREFERENCEMONITORTHATRUNSWITHINOURIBOSKERNELSHADEDREGIONSMAKEUPTHETCB3MAINTAINCOMPATIBILITYWITHCURRENTBROWSERSECURITYPOLICIESOURPRIMARYGOALISTOIMPROVETHEENFORCEMENTOFCURRENTBROWSERPOLICIESWITHOUTCHANGINGCURRENTWEBBASEDAPPLICATIONS4EXPOSEENOUGHBROWSERSTATESANDEVENTSTOENABLENEWBROWSERSECURITYPOLICIESINADDITIONTOENFORCINGCURRENTBROWSERPOLICIES,WEWOULDLIKEOURARCHITECTURETOADAPTEASILYTOFUTUREBROWSERPOLICIES5AVOIDRULEBASEDOSSANDBOXINGFORBROWSERCOMPONENTSFUNDAMENTALLY,RULEBASEDOSSANDBOXINGISABOUTRESTRICTINGUNUSEDOROVERLYPERMISSIVEINTERFACESEXPOSEDBYTODAY’SOPERATINGSYSTEMSHOWEVER,SANDBOXINGSYSTEMSCANBECOMPLEXTHEUBUNTU1004SELINUXREFERENCEPOLICYUSESOVER104KLINESOFPOLICYCODEANDDIFFICULTTOIMPLEMENTCORRECTLY23,51IFOURARCHITECTUREREQUIRESOSSANDBOXINGFORBROWSERCOMPONENTSTHENWESHOULDRETHINKTHEARCHITECTURE22OVERALLARCHITECTUREFIGURE1SHOWSTHEOVERALLIBOSARCHITECTURETHEIBOSARCHITECTUREUSESABASICMICROKERNELAPPROACHWITHATHINKERNELFORMANAGINGHARDWAREANDFACILITATINGMESSAGEPASSINGBETWEENPROCESSESTHESYSTEMINCLUDESUSERMODEDEVICEDRIVERSFORINTERACTINGDIRECTLYWITHHARDWAREDEVICES,SUCHASNETWORKINTERFACECARDSNIC,ANDBROWSERAPIMANAGERSFORACCESSINGTHEDRIVERSANDIMPLEMENTINGBROWSERABSTRACTIONSTHEKEYBROWSERABSTRACTIONSTHATTHEBROWSERAPIMANAGERSIMPLEMENTAREHTTPREQUESTS,COOKIESANDLOCALSTORAGEFORSTORINGPERSISTENTDATA,ANDTABSFORDISPLAYINGUSERINTERFACEUICONTENTWEBAPPSUSETHESEABSTRACTIONSDIRECTLYTOIMPLEMENTBROWSERFUNCTIONALITY,ANDTRADITIONALAPPLICATIONSTRADITIONALAPPSUSEAUNIXLAYERTOACCESSUNIXLIKEABSTRACTIONSONTOPOFTHESEBROWSERABSTRACTIONS221THEIBOSKERNELOURIBOSKERNELISTHESOFTWARETCBFORTHEBROWSERANDINCLUDESRESOURCEMANAGEMENTFUNCTIONALITYANDAREFERENCEMONITORFORSECURITYENFORCEMENTTHEIBOSKERNELALSOHANDLESMANYTRADITIONALOSTASKSSUCHASMANAGINGGLOBALRESOURCES,CREATINGNEWPROCESSES,ANDMANAGINGMEMORYFORAPPLICATIONSTOFACILITATEMESSAGEPASSING,THEIBOSKERNELINCLUDESTHEL4KAPISTACHIO8MESSAGEPASSINGIMPLEMENTATIONANDMMUMANAGEMENTFUNCTIONSALLMESSAGESPASSTHROUGHOURREFERENCEMONITORANDARESUBJECTEDTOOUROVERALLSYSTEMSECURITYPOLICYSECTION4DESCRIBESTHEPOLICIESTHATTHEIBOSKERNELENFORCESANDTHEMECHANISMSITUSESTOIMPLEMENTTHESEPOLICIES222NETWORK,STORAGE,ANDUIMANAGERSTHEIBOSNETWORKSUBSYSTEMHANDLESHTTPREQUESTSANDSOCKETCALLSFORAPPLICATIONSTOHANDLEHTTPREQUESTS,NETWORKPROCESSESCHECKALOCALCACHETOSEEIFTHEREQUESTCANBESERVICEDVIATHECACHE,FETCHANYCOOKIESNEEDEDFORTHEREQUEST,FORMATTHEHTTPDATAINTOATCPSTREAM,ANDTRANSFORMTHATTCPSTREAMINTOASERIESOFETHERNETFRAMESTHATARESENTTOTHENICDRIVERSOCKETNETWORKPROCESSESEXPORTABASICSOCKETAPIANDSIMPLYTRANSFORMTCPSTREAMSTOETHERNETFRAMESFORTRANSMISSIONACROSSTHENETWORKONLYTRADITIONALAPPSCANACCESSOURSOCKETNETWORKPROCESSESTHEIBOSKERNELMANAGESGLOBALSTATES,LIKEPORTALLOCATIONTHEIBOSSTORAGEMANAGERMAINTAINSPERSISTENTSTORAGEFORKEYVALUEDATAPAIRSTHEBROWSERUSESTHESTORAGEMANAGERTOSTOREHTTPCOOKIESANDHTML5LOCALSTORAGEOBJECTS,ANDTHEBASICOBJECTSTOREINCLUDESOPTIONALPARAMETERS,SUCHASPATHANDMAXAGE,TOEXPOSECOOKIEPROPERTIESTOTHEREFERENCEMONITORTHESTORAGEMANAGERUSESSEVERALDIFFERENTNAMESPACESTOISOLATEOBJECTSFROMEACHOTHERWEBAPPSANDNETWORKPROCESSESSHAREANAMESPACEBASEDONTHEORIGINTHETUPLEOFAUNIFORMRESOURCELOCATORTHATTHEYORIGINATEFROM,ANDWEBAPPSANDTRADITIONALAPPSSHAREA“LOCALHOST”NAMESPACE,WHICHISSEPARATEFROMTHEHTTPNAMESPACEALLOTHERDRIVERSANDMANAGERSHAVETHEIROWNPRI3

簡介：FLEXSCFLEXIBLESYSTEMCALLSCHEDULINGWITHEXCEPTIONLESSSYSTEMCALLSLIVIOSOARESUNIVERSITYOFTORONTOMICHAELSTUMMUNIVERSITYOFTORONTOABSTRACTFORTHEPAST30YEARS,SYSTEMCALLSHAVEBEENTHEDEFACTOINTERFACEUSEDBYAPPLICATIONSTOREQUESTSERVICESFROMTHEOPERATINGSYSTEMKERNELSYSTEMCALLSHAVEALMOSTUNIVERSALLYBEENIMPLEMENTEDASASYNCHRONOUSMECHANISM,WHEREASPECIALPROCESSORINSTRUCTIONISUSEDTOYIELDUSERSPACEEXECUTIONTOTHEKERNELINTHEFIRSTPARTOFTHISPAPER,WEEVALUATETHEPERFORMANCEIMPACTOFTRADITIONALSYNCHRONOUSSYSTEMCALLSONSYSTEMINTENSIVEWORKLOADSWESHOWTHATSYNCHRONOUSSYSTEMCALLSNEGATIVELYAFFECTPERFORMANCEINASIGNIFICANTWAY,PRIMARILYBECAUSEOFPIPELINEFLUSHINGANDPOLLUTIONOFKEYPROCESSORSTRUCTURESEG,TLB,DATAANDINSTRUCTIONCACHES,ETCWEPROPOSEANEWMECHANISMFORAPPLICATIONSTOREQUESTSERVICESFROMTHEOPERATINGSYSTEMKERNELEXCEPTIONLESSSYSTEMCALLSTHEYIMPROVEPROCESSOREFFICIENCYBYENABLINGFLEXIBILITYINTHESCHEDULINGOFOPERATINGSYSTEMWORK,WHICHINTURNCANLEADTOSIGNIFICANTLYINCREASEDTEMPORALANDSPACIALLOCALITYOFEXECUTIONINBOTHUSERANDKERNELSPACE,THUSREDUCINGPOLLUTIONEFFECTSONPROCESSORSTRUCTURESEXCEPTIONLESSSYSTEMCALLSAREPARTICULARLYEFFECTIVEONMULTICOREPROCESSORSTHEYPRIMARILYTARGETHIGHLYTHREADEDSERVERAPPLICATIONS,SUCHASWEBSERVERSANDDATABASESERVERSWEPRESENTFLEXSC,ANIMPLEMENTATIONOFEXCEPTIONLESSSYSTEMCALLSINTHELINUXKERNEL,ANDANACCOMPANYINGUSERMODETHREADPACKAGEFLEXSCTHREADS,BINARYCOMPATIBLEWITHPOSIXTHREADS,THATTRANSLATESLEGACYSYNCHRONOUSSYSTEMCALLSINTOEXCEPTIONLESSONESTRANSPARENTLYTOAPPLICATIONSWESHOWHOWFLEXSCIMPROVESPERFORMANCEOFAPACHEBYUPTO116,MYSQLBYUPTO40,ANDBINDBYUPTO105WHILEREQUIRINGNOMODIFICATIONSTOTHEAPPLICATIONS1INTRODUCTIONSYSTEMCALLSARETHEDEFACTOINTERFACETOTHEOPERATINGSYSTEMKERNELTHEYAREUSEDTOREQUESTSERVICESOFFEREDBY,ANDIMPLEMENTEDINTHEOPERATINGSYSTEMKERNELWHILE020004000600080001000012000140001600003050709111315SYSCALLIMPACTONUSERMODEIPCTIMEINCYCLESUSERMODEIPCHIGHERISFASTERSYSCALLEXCEPTIONLOSTPERFORMANCECYCLESFIGURE1USERMODEINSTRUCTIONSPERCYCLESIPCOFXALANFROMSPECCPU2006INRESPONSETOASYSTEMCALLEXCEPTIONEVENT,ASMEASUREDONANINTELCOREI7PROCESSORDIFFERENTOPERATINGSYSTEMSOFFERAVARIETYOFDIFFERENTSERVICES,THEBASICUNDERLYINGSYSTEMCALLMECHANISMHASBEENCOMMONONALLCOMMERCIALMULTIPROCESSEDOPERATINGSYSTEMSFORDECADESSYSTEMCALLINVOCATIONTYPICALLYINVOLVESWRITINGARGUMENTSTOAPPROPRIATEREGISTERSANDTHENISSUINGASPECIALMACHINEINSTRUCTIONTHATRAISESASYNCHRONOUSEXCEPTION,IMMEDIATELYYIELDINGUSERMODEEXECUTIONTOAKERNELMODEEXCEPTIONHANDLERTWOIMPORTANTPROPERTIESOFTHETRADITIONALSYSTEMCALLDESIGNARETHAT1APROCESSOREXCEPTIONISUSEDTOCOMMUNICATEWITHTHEKERNEL,AND2ASYNCHRONOUSEXECUTIONMODELISENFORCED,ASTHEAPPLICATIONEXPECTSTHECOMPLETIONOFTHESYSTEMCALLBEFORERESUMINGUSERMODEEXECUTIONBOTHOFTHESEEFFECTSRESULTINPERFORMANCEINEFFICIENCIESONMODERNPROCESSORSTHEINCREASINGNUMBEROFAVAILABLETRANSISTORSONACHIPMOORE’SLAWHAS,OVERTHEYEARS,LEDTOINCREASINGLYSOPHISTICATEDPROCESSORSTRUCTURES,SUCHASSUPERSCALARANDOUTOFORDEREXECUTIONUNITS,MULTILEVELCACHES,ANDBRANCHPREDICTORSTHESEPROCESSORSTRUCTURESHAVE,INTURN,LEDTOALARGEINCREASEINTHEPERFORMANCEPOTENTIALOFSOFTWARE,BUTATTHESAMETIMETHEREISAWIDENINGGAPBETWEENTHEPERFORMANCEOFEFFICIENTSOFTWAREANDTHEPERFORMANCEOFINEFFICIENTSOFTWARE,PRIMARILYDUETOTHEINCREASINGDISPARITYOFACCESSINGDIFFERENTPROCESSORRESOURCESEG,REGISTERSVSCACHESVSMEMORYSERVERANDSYSTEMINTENSIVEWORKLOADS,WHICHAREOFPARTICULARSYSCALLINSTRUCTIONSCYCLESIPCICACHEDCACHEL2L3DTLBSTAT49721358503732186660255921PREAD37391230003032294679216020PWRITE56893128501850373985316044OPENCLOSE66311916203447240900353428MMAPMUNMAP8977190790474123386939137OPENWRITECLOSE992132815030784811462510549TABLE1SYSTEMCALLFOOTPRINTOFDIFFERENTPROCESSORSTRUCTURESFORTHEPROCESSORSSTRUCTURESCACHESANDTLB,THENUMBERSREPRESENTNUMBEROFENTRIESEVICTEDTHECACHELINEFORTHEPROCESSORISOF64BYTESICACHEANDDCACHEREFERTOTHEINSTRUCTIONANDDATASECTIONSOFTHEL1CACHE,RESPECTIVELYTHEDTLBREPRESENTSTHEDATAPORTIONOFTHETLBKERNELSTACK,CHANGINGTHEPROTECTIONDOMAIN,ANDREDIRECTINGEXECUTIONTOTHEREGISTEREDEXCEPTIONHANDLERSUBSEQUENTLY,RETURNFROMEXCEPTIONISNECESSARYTORESUMEEXECUTIONINUSERMODEWEMEASUREDTHEMODESWITCHTIMEBYIMPLEMENTINGANEWSYSTEMCALL,GETTSCTHATOBTAINSTHETIMESTAMPCOUNTEROFTHEPROCESSORANDIMMEDIATELYRETURNSTOUSERMODEWECREATEDASIMPLEBENCHMARKTHATINVOKEDGETTSC1BILLIONTIMES,RECORDINGTHETIMESTAMPBEFOREANDAFTEREACHCALLTHEDIFFERENCEBETWEENEACHOFTHETHREETIMESTAMPSIDENTIFIESTHENUMBEROFCYCLESNECESSARYTOENTERANDLEAVETHEOPERATINGSYSTEMKERNEL,NAMELY79CYCLESAND71CYCLES,RESPECTIVELYTHETOTALROUNDTRIPTIMEFORTHEGETTSCSYSTEMCALLISMODESTAT150CYCLES,BEINGLESSTHANTHELATENCYOFAMEMORYACCESSTHATMISSESTHEPROCESSORCACHES250CYCLESONOURMACHINE322SYSTEMCALLFOOTPRINTTHEMODESWITCHTIME,HOWEVER,ISONLYPARTOFTHECOSTOFASYSTEMCALLDURINGKERNELMODEEXECUTION,PROCESSORSTRUCTURESINCLUDINGTHEL1DATAANDINSTRUCTIONCACHES,TRANSLATIONLOOKASIDEBUFFERSTLB,BRANCHPREDICTIONTABLES,PREFETCHBUFFERS,ASWELLASLARGERUNIFIEDCACHESL2ANDL3,AREPOPULATEDWITHKERNELSPECIFICSTATETHEREPLACEMENTOFUSERMODEPROCESSORSTATEBYKERNELMODEPROCESSORSTATEISREFERREDTOASTHEPROCESSORSTATEPOLLUTIONCAUSEDBYASYSTEMCALLTOQUANTIFYTHEPOLLUTIONCAUSEDBYSYSTEMCALLS,WEUSEDTHECOREI7HARDWAREPERFORMANCECOUNTERSHPCWERANAHIGHINSTRUCTIONPERCYCLEIPCWORKLOAD,XALAN,FROMTHESPECCPU2006BENCHMARKSUITETHATISKNOWNTOINVOKEFEWSYSTEMCALLSWECONFIGUREDANHPCTOTRIGGERINFREQUENTLYONCEEVERY10MILLIONUSERMODEINSTRUCTIONSSOTHATTHEPROCESSORSTRUCTURESWOULDBEDOMINATEDWITHAPPLICATIONSTATEWETHENSETUPTHEHPCEXCEPTIONHANDLERTOEXECUTESPECIFICSYSTEMCALLS,WHILEMEASURINGTHEREPLACEMENTOFAPPLICATIONSTATEINTHEPROCESSORSTRUCTURESCAUSEDBYKERNELEXECUTIONBUTNOTBYTHEPERFORMANCECOUNTEREXCEPTIONHANDLERITSELF3FORALLEXPERIMENTSPRESENTEDINTHISPAPER,USERMODEAPPLICATIONSEXECUTEIN64BITMODEANDWHENUSINGSYNCHRONOUSSYSTEMCALLS,USETHE“SYSCALL”X8664INSTRUCTION,WHICHISCURRENTLYTHEDEFAULTINLINUXTABLE1SHOWSTHEFOOTPRINTONSEVERALPROCESSORSTRUCTURESFORTHREEDIFFERENTSYSTEMCALLSANDTHREESYSTEMCALLCOMBINATIONSTHEDATASHOWSTHAT,EVENTHOUGHTHENUMBEROFICACHELINESREPLACEDISMODESTBETWEEN2AND5KB,THENUMBEROFDCACHELINESREPLACEDISSIGNIFICANTGIVENTHATTHESIZEOFTHEDCACHEONTHISPROCESSORIS32KB,WESEETHATTHESYSTEMCALLSLISTEDPOLLUTEATLEASTHALFOFTHEDCACHE,ANDALMOSTALLOFTHEDCACHEINTHE“OPENWRITECLOSE”CASETHE64ENTRYFIRSTLEVELDTLBISALSOSIGNIFICANTLYPOLLUTEDBYMOSTSYSTEMCALLSFINALLY,ITISINTERESTINGTONOTETHATTHESYSTEMCALLIMPACTONTHEL2ANDL3CACHESISLARGERTHANONTHEL1CACHES,PRIMARILYBECAUSETHEL2ANDL3CACHESUSEMOREAGGRESSIVEPREFETCHING23SYSTEMCALLIMPACTONUSERIPCULTIMATELY,THEMOSTIMPORTANTMEASUREOFTHEREALCOSTOFSYSTEMCALLSISTHEPERFORMANCEIMPACTONTHEAPPLICATIONTOQUANTIFYTHIS,WEEXECUTEDANEXPERIMENTSIMILARTOTHEONEDESCRIBEDINTHEPREVIOUSSUBSECTIONHOWEVER,INSTEADOFMEASURINGKERNELMODEEVENTS,WEONLYMEASUREDUSERMODEINSTRUCTIONSPERCYCLEIPC,IGNORINGALLKERNELEXECUTIONIDEALLY,USERMODEIPCSHOULDNOTDECREASEASARESULTOFINVOKINGSYSTEMCALLS,SINCETHECYCLESANDINSTRUCTIONSEXECUTEDASPARTOFTHESYSTEMCALLAREIGNOREDINOURMEASUREMENTSINPRACTICE,HOWEVER,USERMODEIPCISAFFECTEDBYTWOSOURCESOFOVERHEADDIRECTTHEPROCESSOREXCEPTIONASSOCIATEDWITHTHESYSTEMCALLINSTRUCTIONTHATFLUSHESTHEPROCESSORPIPELINEINDIRECTSYSTEMCALLPOLLUTIONONTHEPROCESSORSTRUCTURES,ASQUANTIFIEDINTABLE1FIGURES2AND3SHOWTHEDEGRADATIONINUSERMODEIPCWHENRUNNINGXALANFROMSPECCPU2006ANDSPECJBB,RESPECTIVELY,GIVENDIFFERENTFREQUENCIESOFPWRITECALLSTHESEBENCHMARKSWERECHOSENSINCETHEYHAVEBEENCREATEDTOAVOIDSIGNIFICANTUSEOFSYSTEMSERVICES,ANDSHOULDSPENDONLY12OFTIMEEXECUTINGINKERNELMODETHEGRAPHSSHOWTHATDIFFERENTWORKLOADSCANHAVEDIFFERENTSENSITIVITIESTOSYSTEMCALLPOLLUTIONXALANHASABASELINEUSERMODEIPCOF146,BUTTHEIPCDEGRADESBYUPTO65WHENEXECUTINGAPWRITEEVERY1,0002,000INSTRUCTIONS,YIELDINGANIPCBETWEEN058AND050

簡介：EFFICIENTSYSTEMENFORCEDDETERMINISTICPARALLELISMAMITTAIAVIRAM,SHUCHUNWENG,SENHU,BRYANFORDYALEUNIVERSITYABSTRACTDETERMINISTICEXECUTIONOFFERSMANYBENEFITSFORDEBUGGING,FAULTTOLERANCE,ANDSECURITYCURRENTMETHODSOFEXECUTINGPARALLELPROGRAMSDETERMINISTICALLY,HOWEVER,OFTENINCURHIGHCOSTS,ALLOWMISBEHAVEDSOFTWARETODEFEATREPEATABILITY,ANDTRANSFORMTIMEDEPENDENTRACESINTOINPUTORPATHDEPENDENTRACESWITHOUTELIMINATINGTHEMWEINTRODUCEANEWPARALLELPROGRAMMINGMODELADDRESSINGTHESEISSUES,ANDUSEDETERMINATOR,APROOFOFCONCEPTOS,TODEMONSTRATETHEMODEL’SPRACTICALITYDETERMINATOR’SMICROKERNELAPIPROVIDESONLY“SHAREDNOTHING”ADDRESSSPACESANDDETERMINISTICINTERPROCESSCOMMUNICATIONPRIMITIVESTOMAKEEXECUTIONOFALLUNPRIVILEGEDCODEWELLBEHAVEDORNOTPRECISELYREPEATABLEATOPTHISMICROKERNEL,DETERMINATOR’SUSERLEVELRUNTIMEADAPTSOPTIMISTICREPLICATIONTECHNIQUESTOOFFERAPRIVATEWORKSPACEMODELFORBOTHTHREADLEVELANDPROCESSLEVELPARALLELPROGRAMINGTHISMODELAVOIDSTHEINTRODUCTIONOFREAD/WRITEDATARACES,ANDCONVERTSWRITE/WRITERACESINTORELIABLYDETECTEDCONFLICTSCOARSEGRAINEDPARALLELBENCHMARKSPERFORMANDSCALECOMPARABLYTONONDETERMINISTICSYSTEMS,ONBOTHMULTICOREPCSANDACROSSNODESINADISTRIBUTEDCLUSTER1INTRODUCTIONWEOFTENWISHTORUNSOFTWAREDETERMINISTICALLY,SOTHATFROMAGIVENINPUTITALWAYSPRODUCESTHESAMEOUTPUTDETERMINISMISTHEFOUNDATIONOFREPLAYDEBUGGING37,39,46,56,FAULTTOLERANCE15,18,50,ANDACCOUNTABILITYMECHANISMS30,31METHODSOFINTRUSIONANALYSIS22,34ANDTIMINGCHANNELCONTROL4FURTHERASSUMETHESYSTEMCANENFORCEDETERMINISMEVENONMALICIOUSCODEDESIGNEDTOEVADEANALYSISEXECUTINGPARALLELSOFTWAREDETERMINISTICALLYISCHALLENGING,HOWEVER,BECAUSETHREADSSHARINGANADDRESSSPACEORPROCESSESSHARINGRESOURCESSUCHASFILESYSTEMSAREPRONETONONDETERMINISTIC,TIMINGDEPENDENTRACES3,40,42,43USERSPACETECHNIQUESFORPARALLELDETERMINISTICEXECUTION8,10,20,21,44SHOWPROMISEBUTHAVELIMITATIONSFIRST,BYRELYINGONADETERMINISTICSCHEDULERRESIDINGINTHEAPPLICATIONPROCESS,THEYPERMITBUGGYORMALICIOUSAPPLICATIONSTOCOMPROMISEDETERMINISMBYINTERFERINGWITHTHESCHEDULERSECOND,DETERMINISTICSCHEDULERSEMULATECONVENTIONALAPISBYSYNTHESIZINGAREPEATABLEBUTARBITRARYSCHEDULEOFINTERTHREADINTERACTIONS,OFTENUSINGANINSTRUCTIONCOUNTERASANARTIFICIALTIMEMETRICDATARACESREMAIN,THEREFORE,BUTTHEIRMANIFESTATIONDEPENDSSUBTLYONINPUTSANDCODEPATHLENGTHSINSTEADOFON“REAL”TIMETHIRD,THEUSERLEVELINSTRUMENTATIONREQUIREDTOISOLATEANDSCHEDULETHREADS’MEMORYACCESSESCANINCURCONSIDERABLEOVERHEAD,EVENONCOARSEGRAINEDCODETHATSYNCHRONIZESRARELYTOMEETTHESOFTWAREDEVELOPMENT,DEBUGGING,ANDSECURITYCHALLENGESTHATUBIQUITOUSPARALLELISMPRESENTS,ITMAYBEINSUFFICIENTTOSHOEHORNTHESTANDARDNONDETERMINISTICPROGRAMMINGMODELINTOASYNTHETICEXECUTIONSCHEDULEINSTEADWEPROPOSETORETHINKTHEBASICMODELITSELFWEWOULDLIKEAPARALLELENVIRONMENTTHATAIS“DETERMINISTICBYDEFAULT”12,40,EXCEPTWHENWEINJECTNONDETERMINISMEXPLICITLYVIAEXTERNALINPUTSBINTRODUCESNODATARACES,EITHERATTHEMEMORYACCESSLEVEL25,43ORATHIGHERSEMANTICLEVELS3CCANENFORCEDETERMINISMONARBITRARY,COMPROMISEDORMALICIOUSCODEFORSECURITYREASONSANDDISEFFICIENTENOUGHTOUSEFOR“NORMALCASE”EXECUTIONOFDEPLOYEDCODE,NOTJUSTFORINSTRUMENTATIONDURINGDEVELOPMENTASASTEPTOWARDSUCHAMODEL,WEPRESENTDETERMINATOR,APROOFOFCONCEPTOSDESIGNEDAROUNDTHEABOVEGOALSDUETOITSOSLEVELAPPROACH,DETERMINATORSUPPORTSEXISTINGLANGUAGES,CANENFORCEDETERMINISTICEXECUTIONNOTONLYONASINGLEPROCESSBUTONGROUPSOFINTERACTINGPROCESSES,ANDCANPREVENTMALICIOUSUSERLEVELCODEFROMSUBVERTINGTHEKERNEL’SGUARANTEEOFDETERMINISMINORDERTOEXPLORETHEDESIGNSPACEFREELY,DETERMINATORTAKESA“CLEANSLATE”APPROACH,MAKINGFEWCOMPROMISESFORBACKWARDCOMPATIBILITYWITHEXISTINGKERNELSORAPISDETERMINATOR’SPROGRAMMINGMODELCOULDBEIMPLEMENTEDINALEGACYKERNELFORBACKWARDCOMPATIBILITY,HOWEVER,ASPARTOFA“DETERMINISTICSANDBOX”FOREXAMPLE9DETERMINATOR’SUSERLEVELRUNTIMEALSOPROVIDESLIMITEDEMULATIONOFTHEUNIXPROCESS,THREAD,ANDFILEAPIS,TOSIMPLIFYAPPLICATIONPORTINGDETERMINATOR’SKERNELENFORCESDETERMINISMBYDENYINGUSERCODEDIRECTACCESSTOHARDWARERESOURCESWHOSEUSECANYIELDNONDETERMINISTICBEHAVIOR,INCLUDINGREALTIMECLOCKS,CYCLECOUNTERS,ANDWRITABLESHAREDMEMORYDETERMINATORCONSTRAINSUSERCODETORUNWITHINAHIERARCHYOFSINGLETHREADED,PROCESSLIKESPACES,EACHHAVINGAPRIVATEVIRTUALADDRESSSPACETHEKERNEL’SLOWLEVELAPIPROVIDESONLYTHREESYSTEMCALLS,WITHWHICHASPACECANSYNCHRONIZEANDCOMMUNICATEWITHITSIMMEDIATEPARENTANDCHILDRENPOTENTIALLYUSEFULSOURCESOFNONDETERMINISM,SUCHASTIMERS,DETERMINATORENCAPSULATESINTOI/ODEVICES,WHICHUNPRIVILEGEDSPACESCANACCESSCATE,VERIFY,ORANALYZEAPROGRAM’SEXECUTIONHISTORYREPLAYCANBEEFFICIENTWHENONLYI/ONEEDBELOGGED,ASFORAUNIPROCESSORVIRTUALMACHINE22,BUTBECOMESMUCHMORECOSTLYIFINTERNALSOURCESOFNONDETERMINISMDUETOPARALLELISMMUSTALSOBEREPLAYED19,23DETERMINATORTHEREFORETRANSFORMSUSEFULSOURCESOFNONDETERMINISMINTOEXPLICITI/O,WHICHAPPLICATIONSMAYOBTAINVIACONTROLLABLECHANNELS,ANDELIMINATESONLYINTERNALNONDETERMINISMRESULTINGFROMPARALLELISMIFANAPPLICATIONCALLSGETTIMEOFDAY,FOREXAMPLE,THENASUPERVISINGPROCESSCANINTERCEPTTHISI/OTOLOG,REPLAY,ORSYNTHESIZETHESEEXPLICITTIMEINPUTS22ARACEFREEMODELFORSHAREDSTATECONVENTIONALSYSTEMSGIVETHREADSDIRECT,CONCURRENTACCESSTOMANYFORMSOFSHAREDSTATE,SUCHASSHAREDMEMORYANDFILESYSTEMS,YIELDINGDATARACESANDHEISENBUGSIFTHETHREADSFAILTOSYNCHRONIZEPROPERLY25,40,43WHILEREPLAYDEBUGGERS37,39,46,56ANDDETERMINISTICSCHEDULERS8,10,20,21,44MAKEDATARACESREPRODUCIBLEONCETHEYMANIFEST,THEYDONOTCHANGETHEINHERENTLYRACEPRONEMODELINWHICHDEVELOPERSWRITEAPPLICATIONSDETERMINATORREPLACESTHESTANDARDCONCURRENTACCESSMODELWITHAPRIVATEWORKSPACEMODEL,INWHICHDATARACESDONOTARISEINTHEFIRSTPLACETHISMODELGIVESEACHTHREADACOMPLETE,PRIVATEVIRTUALREPLICAOFALLLOGICALLYSHAREDSTATEATHREADMAYACCESS,INCLUDINGSHAREDMEMORYANDFILESYSTEMSTATEATHREAD’SNORMALREADSANDWRITESAFFECTONLYITSPRIVATEWORKINGSTATE,ANDDONOTINTERACTDIRECTLYWITHOTHERTHREADSINSTEAD,DETERMINATORACCUMULATESEACHTHREADS’SCHANGESTOSHAREDSTATE,THENRECONCILESTHESECHANGESAMONGTHREADSONLYATPROGRAMDEFINEDSYNCHRONIZATIONPOINTSTHISMODELISRELATEDTOANDINSPIREDBYEARLYPARALLELFORTRANSYSTEMS7,51,REPLICATEDFILESYSTEMS47,TRANSACTIONALMEMORY33,52ANDOPERATINGSYSTEMS48,ANDDISTRIBUTEDVERSIONCONTROLSYSTEMS29,BUTTOOURKNOWLEDGEDETERMINATORISTHEFIRSTOSTOINTRODUCEAMODELFORPERVASIVETHREADANDPROCESSLEVELDETERMINISMIFONETHREADEXECUTESTHEASSIGNMENT‘XY’WHILEANOTHERCONCURRENTLYEXECUTES‘YX’,FOREXAMPLE,THESEASSIGNMENTSRACEINTHECONVENTIONALMODEL,BUTARERACEFREEUNDERDETERMINATORANDALWAYSSWAPXWITHYEACHTHREAD’SREADOFXORYALWAYSSEESTHE“OLD”VERSIONOFTHATVARIABLE,SAVEDINTHETHREAD’SPRIVATEWORKSPACEATTHELASTEXPLICITSYNCHRONIZATIONPOINTFIGURE1ILLUSTRATESAMOREREALISTICEXAMPLEOFAGAMEORSIMULATOR,WHICHUSESANARRAYOF“ACTORS”PLAYERS,PARTICLES,ETCTOREPRESENTSOMELOGICAL“UNIVERSE,”ANDUPDATESALLOFTHEACTORSINPARALLELATEACHTIMESTEPTOUPDATETHEACTORS,THEMAINTHREADFORKSACHILDTHREADTOPROCESSEACHACTOR,THENSYNCHRONIZESBYJOININGALLTHESECHILDTHREADSTHECHILDTHREADCODETOUPDATEEACHACTORISSHOWN“INLINE”WITHINTHEMAINFUNCTION,WHICHSTRUCTACTORSTATEACTORNACTORSMAININITIALIZEALLELEMENTSOFACTORARRAYFORTIME0TIMEFORI0INACTORSIIFTHREADFORKIINCHILD//CHILDTHREADTOPROCESSACTORIEXAMINESTATEOFNEARBYACTORSUPDATESTATEOFACTORIACCORDINGLYTHREADEXITFORI0INACTORSITHREADJOINIFIGURE1CPSEUDOCODEFORLOCKSTEPTIMESIMULATION,WHICHCONTAINSADATARACEINSTANDARDCONCURRENCYMODELSBUTISBUGFREEUNDERDETERMINATORUNDERUNIXWORKSONLYWITHPROCESSLEVELFORKDETERMINATOROFFERSTHISCONVENIENCEFORSHAREDMEMORYTHREADSASWELL,ASDISCUSSEDLATERINSECTION44INTHISEXAMPLE,EACHCHILDTHREADREADSTHE“PRIOR”STATEOFANYORALLACTORSINTHEARRAY,THENUPDATESTHESTATEOFITSASSIGNEDACTOR“INPLACE,”WITHOUTANYEXPLICITCOPYINGORADDITIONALSYNCHRONIZATIONWITHSTANDARDTHREADSTHISCODEHASAREAD/WRITERACEEACHCHILDTHREADMAYSEEANARBITRARYMIXOF“OLD”AND“NEW”STATESASITEXAMINESOTHERACTORSINTHEARRAYUNDERDETERMINATOR,HOWEVER,THISCODEISCORRECTANDRACEFREEEACHCHILDTHREADREADSONLYITSPRIVATEWORKINGCOPYOFTHEACTORSARRAY,WHICHISUNTOUCHEDEXCEPTBYTHECHILDTHREADITSELFSINCETHEMAINTHREADFORKEDTHATCHILDASTHEMAINTHREADREJOINSALLITSCHILDTHREADS,DETERMINATORMERGESEACHCHILD’SACTORARRAYUPDATESBACKINTOTHEMAINTHREAD’SWORKINGCOPY,FORUSEINTHENEXTTIMESTEPWHILEREAD/WRITERACESDISAPPEARINDETERMINATOR’SMODEL,TRADITIONALWRITE/WRITERACESBECOMECONFLICTSIFTWOCHILDTHREADSCONCURRENTLYWRITETOTHESAMEACTORARRAYELEMENT,FOREXAMPLE,DETERMINATORDETECTSTHISCONFLICTANDSIGNALSARUNTIMEEXCEPTIONWHENTHEMAINTHREADATTEMPTSTOJOINTHESECONDCONFLICTINGCHILDINTHECONVENTIONALMODEL,BYCONTRAST,THETHREADS’EXECUTIONSCHEDULESMIGHTCAUSEEITHEROFTHETWOWRITESTO“WIN”ANDSILENTLYPROPAGATEITSLIKELYERRONEOUSVALUETHROUGHOUTTHECOMPUTATIONRUNNINGTHISCODEUNDERACONVENTIONALDETERMINISTICSCHEDULERCAUSESTHE“WINNER”TOBEDECIDEDBASEDONASYNTHETIC,REPRODUCIBLETIMEMETRICEG,INSTRUCTIONCOUNTRATHERTHANREALTIME,BUTTHERACEREMAINSANDMAYSTILLMANIFESTORVANISHDUETOSLIGHTCHANGESININPUTSORINSTRUCTIONPATHLENGTHS23ARACEFREESYNCHRONIZATIONAPICONVENTIONALTHREADSCANSTILLBEHAVENONDETERMINISTICALLYEVENINACORRECTLYLOCKEDPROGRAMWITHNOLOW

簡介：；．’’．．分類號Ｒ２７１．１９密級公開，＾－單位代碼；１０７６０北＾學號１０７加２１２６４９４＾＾滬．－．Ａ亡＇．＇＇Ｖ＇，Ｖ，，．一．，－．，＇－Ｊ．，古．‘Ｖ．Ｒ－．一．．；＇．．．Ｖ＇．．．、．．‘－‘＇譜斯壤處種乂營ＸＩＮＪＩＡＮＧＭＥＤＩＣＡＬＵＮＩＶＥＲＳＩＴＹ碩±學位論文ＴＨＥＳＩＳＯＦＭ乂ＳＴＥＲＤＥＧＲＥＥ臨床醫(yī)學專業(yè)學位（學歷教育）論文題目圍絕鐘期綜合征與西北燥證的相關性調查研究研究生李紫營，置指導教師劉愛玲副教授主任醫(yī)師．，＾專業(yè)學位領域中醫(yī)婦科學總皆扣＞Ｆ．＇ＶＲ．‘Ｖ令研究方向中醫(yī)婦科病防治鉛Ｖ。．－，＇研究起止時間２〇１４．〇１２〇１５．〇１．茄違戸＾’．所在學院中醫(yī)學院’繫；項２０１５年０３月．＿，、，社，＜Ｔ＇－Ｓ－．＇．‘；；．．．－■．ＩＦ；；’產＇斯．．＇＇柏祭了．／；．；＇Ｋ．Ｎ＇廣游貨．午’驟ＫＲ－圍絕經期綜合征與西北燥證的相關性調查研圍絕經期綜合征與西北燥證的相關性調查研究2015年03月研究生研究生李紫萱李紫萱指導教師指導教師劉愛玲劉愛玲副教授教授主任醫(yī)師主任醫(yī)師專業(yè)學位領域專業(yè)學位領域中醫(yī)婦科學婦科學研究方向研究方向中醫(yī)婦科病防治中醫(yī)婦科病防治